--- v4 ---
Improved fantasy awareness
General Improvements
Minor bug fixes
--- v3 ---
Improved skin textures
Better environment adherence
Eye Improvements
Shadow and Light improvements.
--- v2 ---
Improved prompt adherence (especially for non-human requests)
Better animal generation
More refined textures
Eye improvements
Lighting Improvements
--- v1 ---
QuadPipe is an incredibly responsive photo realistic model. It maintains native tagging while enhancing the quality of the output to a new level. QuadPipe handles text about as well as possible for SDXL and seems determined to get it close.
This is my best work. I've never created a model that performs better than this one.
I will continue to update this, though on a slower scale, to make sure that things are very refined before release. As always, let me know if you find any bugs, and I'll incorporate your discoveries into my next build. Thank you all for your support, I couldn't have created this model without your faith in me.
Check my prompts to learn about settings etc... but in general this is what I was using when I created the demos:
768x1152
CFG 5.5-9.5
4x_foolhardy_Remacri (upscaler)
30 hires steps
0.27 Denoising Strength
Upscale by 1.5
Sampler: DPM++ 2M SDE Heun
Schedule: Karras
Sampling Steps: 50
Description
FAQ
Comments (9)
So far, comparing against V3, the general details on v4 are so good , the only problems that I'm currently facing are for example actions with hands or some nsfw actions like deepthroating blowjob ,other than that, the prompt adherence I would say improved against V3 👌🏻
First, thank you for the kind words!
Next, I don't test against NSFW typically, so I suppose there are bound to be some issues that pop up. Largely though, issues that appear there will likely appear elsewhere sometimes. I'll take a look at hands and see if I can work to improve them more. Hands (as always) are a pain in the butt, but I've made a lot of progress there and I continue to try to refine them.
@QuadPipe thank you, I imagine that straight nsfw is not your main point, although I did some crazy stuff that is awesome haha, but hands are actually good tho, I just meant for example, actions that require the use of hands,not the hand itself, but so far is the only problem,in any other aspects it's a truly successor of v3,the realism is awesome
what models was this merged from
It's not a model merge. It does begin with the Core SDXL model and then it uses independently trained and fine-tuned LoRAs that are merged in to nudge it toward where it is today. Version 2 begins with Version 1, and then fine-tuning. Version 3 begins with Version 2 etc...
So this isn't a "Model Merge" even though it uses LoRA merges that I train here to adjust the direction.
great job ! Bravo
does it work in MFLUX-WEBUI ?
I haven't tested it in MFLUX... I actually don't think I even knew about MFLUX. I'd be eager to know how it goes.
would love to see more examples containing text
Sure, I'll post some later today/tomorrow. It does a reasonable job of text for an SDXL model. I think my model, MegaChonk probably does a better job with text - but there was a lot of text-specific training with that one. I would say that with any SDXL model, it takes a few generations to get there - depending on what you're trying to say. Sometimes this one hits it right away, other times it takes 10-15 generations to get close. It's not up to FLUX text standards by any means. - I'll keep refining it though.